Amiibo Theory: Adaptation

sarcasmspasm
Don't talk much, huh?

Posts: 7

Amiibo Theory: Adaptation Feb 25, 2016 13:17:35 GMT via mobile Cloud and map like this

Post by sarcasmspasm on Feb 25, 2016 13:17:35 GMT

Hey All!!! This is my first thread ever created (besides my intro). I hope you like it.

I am constantly testing what is the best method that amiibo will learn from. While I have nowhere near tried all of my theories, I wish to share some of my preliminary findings.

I've noticed many amiibo fights where the victor was down substantially, but made a near flawless comeback. This leads me to believe that the true key to victory is adaptation. Therefore, I set out to find what is the best way to teach your amiibo to adapt. Below is my experiment, conclusion and application.

EXPERIMENT
To keep this experiment untainted by human hands, I decided to train these amiibo by cpu. This way they would not incur the bad habits I or my trained amiibo may accidentally teach it. This was not a test of skill, but rather, it was a test of how they learn. I would then put these amiibo against each other in three 10-stock matches and ten 2-stock matches to determine how good their adapting skills and tourney-style short game were.

Note: In between these matches, I left the screen and did not save either amiibo. I feel this was a better test to show whose AI had grown better.

The contenders were two vanilla Bowser amiibo. (Named Shelldon and King Koopa.)

I trained Shelldon firstly by two simple 99-stock matches against a level-9 cpu. He died rapidly in the beginning, but greatly closed the gap towards his later lives of the first match. The ending score was 99-49 KOs. (Cpu won.) This left Shelldon at level 46. During the second match,the roles were reversed. Shelldon gradually destroyed the computer, before making faster, more aggressive KOs. Ending score was 99-41. (Shelldon)

I began King Koopa's training with two 99-stock matches; one with a level 1 cpu, the other with a level 9. My reasoning was to see if forming its own move set by having a rag doll to kick, then sharpening those moves against a better defense, would produce better results. KK was aggressive in the first match where he won 99-17, but when he fought the level nine, that aggression hit a brick wall because he was being punished for his lack of refinement. KK still won with 99-57. However, at the end of BOTH matches, KK was only level 48!!!

Now I made them fight each other. Though I stupidly forgot to record the amiibo fights, I will give you my best description. Shelldon came more defensively while KK came more aggressively. Shelldon won all three 10-stock matches. KK surprisingly won 6/10 2-stock matches. In light of this, I decided to firther research with lower level cpu training.

I reset KK and sent him in 22-stock matches (The first match was 20 so the total of stocks would remain 198.) against level 1, 2, 3, etc. He won all of the fights while only losing about three stocks each match. He hit level 50 during the level 8 fight.

I faced KK and Shelldon off again... The results were beyond thrilling. KK DEMOLISHED Shelldon in EVERY match!!! KK was aggressive, but he had much more defensive skills then before.

Now that KK was the one to beat, I reset Shelldon and repeated the process I used previously, but with eleven 2-stock matches for each level. (Ten for level 1.) Shelldon won every match except one (his shield was broken at the edge.). Approximately 85% of his matches were 2-0. He hit level 50 during the last level 7 match.

I finally (with great anticipation,) did my final face off between the two. The results were not as satisfactory as I had hoped. KK won all three 10-stock matches and 7/10 2-stock matches.

Experiment complete.

CONCLUSION
I feel the best way to train amiibo is to face them against an opponent at the same level as it. If not the same, better. Do your best NOT to let them fight weaker opponents.

Shelldon was forced to adapt when he realized level 9 was much better than he.
KK rested on his laurels too much and his learning was stunted.

KK 2.0 was a different story. He was overly aggressive due to killing so much and dying so little, but he was facing a computer that was more on his level during his training than either of the two before him. This allowed him to adapt more naturally than Shelldon. Since he fought opponents closer to him in skill, he took their defensive techniques. A true masterpiece of adaptation.

Shelldon 2.0 had 10x the match experience as KK 2.0, but I feel he couldn't keep up with him because most of his matches were landslide victories. He didn't learn from those matches due to their brevity and lack of challenge.

All of this is wonderful, (to me, at least.) but it doesn't mean anything if you don't learn anything from it...

APPLICATION
Since I feel that the best way is to set up amiibo against an opponent with the same skill level, what do you do? Simple. Set handicaps. (Are you kidding me Spasm?!?! All of that for that?!?!) This will allow amiibo (I couldn't get through this without asking this question... Why is the plural of amiibo still amiibo?) of varying skill levels to adapt to their opponent. I theorize that the result will be greatly improved skill levels for both parties.

If you wish to know how to set the handicap, set the two amiibo to a 10-stock match. On the victory screen, view the winner's damage taken. Then do as follows...

X=winner's damage taken y=winner's lives lost
z=x/y a= winner's lives remaining b=z*a
c=10 (or the number of stocks in the match.)

x/y=z z*a=b b/c=handicap

...I said in my intro thread that I'm rather calculating.

? Let me give an example.

Little Mac wins with two lives left...

Damage taken 2000

Lives lost 8

Lives remaining 2

2000/8=250

250*2=500

500/10=50

So set Little Mac's handicap to 50% and send him against this other amiibo for a thirty minute match. (It could be longer or maybe a stock match.)

Afterwards, send team in another 10-stock match.

MAKE SURE YOU COMPARE HOW GOOD THE TWO FIGHTS ARE!!!

Ask yourself: Do they still fight the same? Was the fight close? What seems better? What seems worse?

I would LOVE to hear from your results. Please post them if you think this experiment is worth trying.

Thanks SO much for taking the time to read this!!! (Long winded as it was. Sorry if I sounded confusing. (It made sense in my head.))

??

Love,
Sarcasm Spasm

sarcasmspasm
Don't talk much, huh?

Posts: 7

Amiibo Theory: Adaptation Feb 25, 2016 13:22:39 GMT via mobile

Post by sarcasmspasm on Feb 25, 2016 13:22:39 GMT

One last note: If you guys have any other crazy amiibo theories or experiments, please let me know!!! I get a kick out of them.

ANOTHER final note: (Sorry.) If your handicap doesn't seem right, use your gut to think of a good handicap. (One time I calculated a 200% handicap for my Little Mac. It didn't seem like a good idea, but out of curiosity, I did it anyway. He lost by about 40 stocks. On the other hand, my fox amiibo DID learn explosive perfect shield and used it like crazy afterwards.) Anyway, listen to your heart. (Listening to your head makes you post threads like the one above.) =p

Last Edit: Feb 25, 2016 15:16:05 GMT by Cloud

EPM
Quiet Member

I seek battle, not badges. Pride, not prizes. Mastery, not victory.

Posts: 35

Amiibo Theory: Adaptation Feb 25, 2016 15:47:02 GMT

Post by EPM on Feb 25, 2016 15:47:02 GMT

I adore theories of amiibo "brain structure", so greatly enjoyed this read. It's worth noting that your experiment confirms what the game's tips say themselves: amiibo learn more from losing than they do by winning.

I've always operated on the assumption of reinforcement, as virtually every guide and trainer (and direct experience) will confirm that amiibo use what works for them and avoid what doesn't. The way they get 'stuck' on things is because they can't really innovate past the occasional random experiment, and if a strategy doesn't net them obvious/immediate benefit (such as fully charging up a move that can be charged vs. spamming basic shots) it's harder to teach them. Likewise, that's how they get stuck spamming moves that work for them, especially grabs.

One experiment I saw on the Smashboards forum comes to mind. One amiibo was trained up to level 50, and then the second, identical amiibo was trained to level 50 exclusively by mirror-matching the "elder sibling". The result was that the second amiibo was quite a bit smarter, but attempts at 3rd and 4th generations (by resetting the first amiibo) produced weaker subjects because the 1st/eldest amiibo's human contact had been diluted out. So the resulting amiibo were over-specialized at beating their mirror-match siblings, but inept at anything else.

A tantalizing bit of trivia came to me by analyzing the actual NFC data structure: When an amiibo is erased, the data isn't cleared, it's just marked as empty and write-able. I've experimented with a Mii Brawler amiibo that never impressed me, and have seen a few mentions here and there across the internet, to see if a fresh level 50 amiibo being reset to level 1 and then retrained immediately won't retain some of its good habits. The results have been inconclusive, but next time I get new amiibo, I'm going to 0-50 train it as best I can, reset it, and then repeat that several times.

(Also, fish and sheep are plural the same way amiibo are, if that helps you.)

M.2 (Mewtwo), Rowsbowser (Bowser), Pikadorf (Ganondorf), Lucina (Marth), Shart (Shulk), HubStyle (MegaMan), Furbait (Lucario), Tajador (Captain Falcon), Sponge (Luigi), Lil' Effort (Little Mac), Dr. Regal (Mii Brawler).

Havok
Full Member

Posts: 152

Amiibo Theory: Adaptation Feb 25, 2016 16:02:56 GMT

Post by Havok on Feb 25, 2016 16:02:56 GMT

Pretty good read and an interesting experiment.

Tourney Entries: Villager (AWT5), Charizard (AWT6) [Top 8], Ryu (vanilla)(AWT7), Lucario (WAS0) [3rd place], Marth (AM1) [4th place], Villager (WAS1), Bowser (AM2), Villager (AASC), Lucario (WAS2), Lucario (CoC0) [Top 6], Marth (WAS3) [Top 8], Gannondorf (AM3), Marth (AM3), Bowser (AASC2) [3rd place], Diddy Kong (WAS4), Lucario (CoC1), Marth (WAS5) [Top 6], Diddy Kong (AASC3), Charizard (WAS6), Mewtwo (WAS7), Lucario (AASC4), Ryu (Vanilla) (WAS8), **Mystery Team** (WAS9)

sarcasmspasm Don't talk much, huh? Posts: 7	Amiibo Theory: Adaptation Feb 25, 2016 16:32:39 GMT via mobile Select Post Deselect Post Link to Post Member Give Gift Back to Top Post by sarcasmspasm on Feb 25, 2016 16:32:39 GMT Generational learning...? I have a new theory!!! I'll keep you guys POSTed!!! =p

keethkon
Quiet Member

Posts: 49

Amiibo Theory: Adaptation Feb 26, 2016 3:44:49 GMT via mobile Cloud likes this

Post by keethkon on Feb 26, 2016 3:44:49 GMT

Amiibos tend to learn from who they fight against when more time is spent putting them in the same stages over and over again to fight each other. Also, reseting amiibos constantly is not the best way to unlock the true potential each of them posess in fights because they learn overtime after level 50. The more time someone spends on their amiibos, the better they become when they are put into tournaments.

EPM
Quiet Member

I seek battle, not badges. Pride, not prizes. Mastery, not victory.

Posts: 35

Amiibo Theory: Adaptation Feb 27, 2016 20:54:05 GMT Cloud likes this

Post by EPM on Feb 27, 2016 20:54:05 GMT

I like to think of Amiibo decision making as a Plinko board:

Experience and training is all about adjusting the "pegs" (that is, weightable factors into what decision the amiibo ultimately makes) on the board, or walling off certain trajectories. This is a decision-tree, basically, but easy to visualize. Experience is the ultimate deciding factor for amiibo competence, I'm certain. I wish I had a way to tell how many battles mine have fought, but lack such means. A "bigger board" of more experiences offers a greater diversity of considerations, and also arranges the pegs to favor some outcomes over others based on immediate circumstances.

An excellent example is Mewtwo's Shadow Ball. Spending the time to charge it, rather than peppering a foe with weaker shots to control the neutral game, is something one needs to teach the amiibo early on. Indeed, I did not use uncharged Shadow Balls once in the course of its 1-50 period. As Mewtwo grew exposed to other amiibo, however, it learned to space with uncharged Shadow Balls anyway, which led to a frustrating tug-of-war of influences on its habits. You see, an amiibo can't really plan ahead. Only computers much, much more powerful than the processing space allocated to the amiibo AI could actually extrapolate current conditions and then prepare accurately. So from an amiibo perspective, what advantage does charging Shadow Ball offer? The opponent is able to draw in closer, it leaves them immobilized and vulnerable, and even if shielding or dodging out of the charge is feasible, it still puts them in an unfavorable situation. And a charged shot is an all-your-eggs-in-one-basket attack.

Over time, however, my mirror matches drilled in the idea of charging up Shadow Ball when at a safe distance (the minimum seems to be 3/4 of Final Destination's length), immediately upon scoring a KO, or via respawn invulnerability. The latter two are inconsistent, but happen increasingly often, so I can definitely say they're being picked up on. One very clear case of adaptation, though, is that my Mewtwo amiibo will almost always charge up a Shadow Ball when I myself do. This is because, obviously, I'm not a threat to it while I do so, and it's learned I have a habit of fully charging. This lead to mutually reinforced habits, as I would fully charge while it did. All of our mirror matches now start the same way unless I deliberate break pattern: both of us fully charging up a Shadow Ball.

So ultimately, experience taught Mewtwo the idea of how far is 'safe distance' to charge SB, and if it's any closer, it will space the opponent with uncharged shots to zone them. A favorite trick I taught it was to pepper shots while slowly letting the gap close, and then dash-grabbing to either get the opponent while they're shielding or in hitstun.

One recent training method I've gone with is "bulk input". Namely, exposing the amiibo to as much as possible, and letting it filter through that new information over the course of many matches. Afterwards, I then battle the amiibo to evaluate what it's learned, and assist in 'sculpting' any changes in behavior.

Interestingly enough, my amiibo have consistently been more "quick-learning" immediately after a stat or equipment bonus shift, such as Mewtwo picking up D-tilt -> U-tilt -> U-Air only after I'd given it an experimental stat reallocation. Speed stats in particular seem to be the most important for causing an amiibo to re-evaluate their behaviors and become more plastic-minded, as their mobility influences the "what they can do right now" far more than attack or defense, which dictate "how well it works", loosely speaking.